How AI and Human Behaviors Shape Psychosocial Effects of Extended Chatbot Use: A Longitudinal Randomized Controlled Study

TL;DR

This study investigates how AI chatbot design and user behavior shape psychosocial outcomes over four weeks using a 3×3 randomized design across three modalities (text, neutral, engaging voice) and three conversation tasks (open-ended, non-personal, personal). Despite few main effects on loneliness, real-world socialization, emotional dependence, or problematic AI use, longer voluntary usage was consistently linked to worse outcomes, and personal conversations modestly reduced dependence and problematic use. Exploratory analyses reveal that user perceptions and traits—such as trust in AI, perceived friendship, and prior chatbot experience—predict higher dependence and problematic use, while higher anthropomorphism and emotional content in text interactions influence dynamics differently across modalities. The work highlights the complexity of extending AI companionship and the potential of duration-focused signals as intervention levers to mitigate negative effects. Overall, a holistic view of model behavior, user engagement, and personality plays a crucial role in shaping AI-assisted psychosocial experiences.

Abstract

As people increasingly seek emotional support and companionship from AI chatbots, understanding how such interactions impact mental well-being becomes critical. We conducted a four-week randomized controlled experiment (n=981, >300k messages) to investigate how interaction modes (text, neutral voice, and engaging voice) and conversation types (open-ended, non-personal, and personal) influence four psychosocial outcomes: loneliness, social interaction with real people, emotional dependence on AI, and problematic AI usage. No significant effects were detected from experimental conditions, despite conversation analyses revealing differences in AI and human behavioral patterns across the conditions. Instead, participants who voluntarily used the chatbot more, regardless of assigned condition, showed consistently worse outcomes. Individuals' characteristics, such as higher trust and social attraction towards the AI chatbot, are associated with higher emotional dependence and problematic use. These findings raise deeper questions about how artificial companions may reshape the ways people seek, sustain, and substitute human connections.

Figures (13)

• Figure 1: Conceptual framework of the study. The study examines how different interaction modalities and conversation tasks influence user's psychosocial outcomes over a four-week period. The study explores how user behavior, human perception of AI and model behavior impact psychosocial outcomes including loneliness, socialization with people, emotional dependence on AI, and problematic use of AI.
• Figure 2: Changes in psychosocial outcomes over the 4-week study duration. Each point represents one observation. Lines represent changes in the mean values. Shaded areas represent standard errors.
• Figure 3: Point plots of regression results for final psychosocial outcomes for text, neutral voice, and engaging voice modalities. Scales: Loneliness (1-4); Socialization with people (0-5); Emotional dependence (1-5); Problematic use of the chatbot (1-5). Error bar: standard error.
• Figure 4: Point plots of regression results for the final psychosocial outcomes for open-ended, non-personal, and personal conversation topics. Scales: Loneliness (1-4); Socialization with people (0-5); Emotional dependence (1-5); Problematic use of the chatbot (1-5). Error bar: standard error.
• Figure 5: Amount of daily time spent (duration) with the chatbot across conditions. (A) Each point represents the average daily duration for each day with a trend line with a shaded confidence interval. (B) Distribution of daily duration per participant. Dashed line represents the mean. (C) Daily duration per participant grouped by modality. (D) Daily duration per participant, grouped by Task. **: p$<$0.01, ***: p$<$0.001. Error bars represent standard error.